xref: /linux/fs/nfs/blocklayout/blocklayout.c (revision de008c9ba5684f14e83bcf86cd45fb0e4e6c4d82)
1 /*
2  *  linux/fs/nfs/blocklayout/blocklayout.c
3  *
4  *  Module for the NFSv4.1 pNFS block layout driver.
5  *
6  *  Copyright (c) 2006 The Regents of the University of Michigan.
7  *  All rights reserved.
8  *
9  *  Andy Adamson <andros@citi.umich.edu>
10  *  Fred Isaman <iisaman@umich.edu>
11  *
12  * permission is granted to use, copy, create derivative works and
13  * redistribute this software and such derivative works for any purpose,
14  * so long as the name of the university of michigan is not used in
15  * any advertising or publicity pertaining to the use or distribution
16  * of this software without specific, written prior authorization.  if
17  * the above copyright notice or any other identification of the
18  * university of michigan is included in any copy of any portion of
19  * this software, then the disclaimer below must also be included.
20  *
21  * this software is provided as is, without representation from the
22  * university of michigan as to its fitness for any purpose, and without
23  * warranty by the university of michigan of any kind, either express
24  * or implied, including without limitation the implied warranties of
25  * merchantability and fitness for a particular purpose.  the regents
26  * of the university of michigan shall not be liable for any damages,
27  * including special, indirect, incidental, or consequential damages,
28  * with respect to any claim arising out or in connection with the use
29  * of the software, even if it has been or is hereafter advised of the
30  * possibility of such damages.
31  */
32 
33 #include <linux/module.h>
34 #include <linux/init.h>
35 #include <linux/mount.h>
36 #include <linux/namei.h>
37 #include <linux/bio.h>		/* struct bio */
38 #include <linux/prefetch.h>
39 
40 #include "../pnfs.h"
41 #include "../nfs4session.h"
42 #include "../internal.h"
43 #include "blocklayout.h"
44 
45 #define NFSDBG_FACILITY	NFSDBG_PNFS_LD
46 
47 MODULE_LICENSE("GPL");
48 MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>");
49 MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver");
50 
51 static bool is_hole(struct pnfs_block_extent *be)
52 {
53 	switch (be->be_state) {
54 	case PNFS_BLOCK_NONE_DATA:
55 		return true;
56 	case PNFS_BLOCK_INVALID_DATA:
57 		return be->be_tag ? false : true;
58 	default:
59 		return false;
60 	}
61 }
62 
63 /* The data we are handed might be spread across several bios.  We need
64  * to track when the last one is finished.
65  */
66 struct parallel_io {
67 	struct kref refcnt;
68 	void (*pnfs_callback) (void *data);
69 	void *data;
70 };
71 
72 static inline struct parallel_io *alloc_parallel(void *data)
73 {
74 	struct parallel_io *rv;
75 
76 	rv = kmalloc_obj(*rv, GFP_NOFS);
77 	if (rv) {
78 		rv->data = data;
79 		kref_init(&rv->refcnt);
80 	}
81 	return rv;
82 }
83 
84 static inline void get_parallel(struct parallel_io *p)
85 {
86 	kref_get(&p->refcnt);
87 }
88 
89 static void destroy_parallel(struct kref *kref)
90 {
91 	struct parallel_io *p = container_of(kref, struct parallel_io, refcnt);
92 
93 	dprintk("%s enter\n", __func__);
94 	p->pnfs_callback(p->data);
95 	kfree(p);
96 }
97 
98 static inline void put_parallel(struct parallel_io *p)
99 {
100 	kref_put(&p->refcnt, destroy_parallel);
101 }
102 
103 static struct bio *
104 bl_submit_bio(struct bio *bio)
105 {
106 	if (bio) {
107 		get_parallel(bio->bi_private);
108 		dprintk("%s submitting %s bio %u@%llu\n", __func__,
109 			bio_op(bio) == READ ? "read" : "write",
110 			bio->bi_iter.bi_size,
111 			(unsigned long long)bio->bi_iter.bi_sector);
112 		submit_bio(bio);
113 	}
114 	return NULL;
115 }
116 
117 static bool offset_in_map(u64 offset, struct pnfs_block_dev_map *map)
118 {
119 	return offset >= map->start && offset < map->start + map->len;
120 }
121 
122 static struct bio *
123 do_add_page_to_bio(struct bio *bio, int npg, enum req_op op, sector_t isect,
124 		struct page *page, struct pnfs_block_dev_map *map,
125 		struct pnfs_block_extent *be, bio_end_io_t end_io,
126 		struct parallel_io *par, unsigned int offset, int *len)
127 {
128 	struct pnfs_block_dev *dev =
129 		container_of(be->be_device, struct pnfs_block_dev, node);
130 	u64 disk_addr, end;
131 
132 	dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
133 		npg, (__force u32)op, (unsigned long long)isect, offset, *len);
134 
135 	/* translate to device offset */
136 	isect += be->be_v_offset;
137 	isect -= be->be_f_offset;
138 
139 	/* translate to physical disk offset */
140 	disk_addr = (u64)isect << SECTOR_SHIFT;
141 	if (!offset_in_map(disk_addr, map)) {
142 		if (!dev->map(dev, disk_addr, map) || !offset_in_map(disk_addr, map))
143 			return ERR_PTR(-EIO);
144 		bio = bl_submit_bio(bio);
145 	}
146 	disk_addr += map->disk_offset;
147 	disk_addr -= map->start;
148 
149 	/* limit length to what the device mapping allows */
150 	end = disk_addr + *len;
151 	if (end >= map->disk_offset + map->len)
152 		*len = map->disk_offset + map->len - disk_addr;
153 
154 retry:
155 	if (!bio) {
156 		bio = bio_alloc(map->bdev, bio_max_segs(npg), op, GFP_NOIO);
157 		bio->bi_iter.bi_sector = disk_addr >> SECTOR_SHIFT;
158 		bio->bi_end_io = end_io;
159 		bio->bi_private = par;
160 	}
161 	if (bio_add_page(bio, page, *len, offset) < *len) {
162 		bio = bl_submit_bio(bio);
163 		goto retry;
164 	}
165 	return bio;
166 }
167 
168 static void bl_mark_devices_unavailable(struct nfs_pgio_header *header, bool rw)
169 {
170 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
171 	size_t bytes_left = header->args.count;
172 	sector_t isect, extent_length = 0;
173 	struct pnfs_block_extent be;
174 
175 	isect = header->args.offset >> SECTOR_SHIFT;
176 	bytes_left += header->args.offset - (isect << SECTOR_SHIFT);
177 
178 	while (bytes_left > 0) {
179 		if (!ext_tree_lookup(bl, isect, &be, rw))
180 				return;
181 		extent_length = be.be_length - (isect - be.be_f_offset);
182 		nfs4_mark_deviceid_unavailable(be.be_device);
183 		isect += extent_length;
184 		if (bytes_left > extent_length << SECTOR_SHIFT)
185 			bytes_left -= extent_length << SECTOR_SHIFT;
186 		else
187 			bytes_left = 0;
188 	}
189 }
190 
191 static void bl_end_io_read(struct bio *bio)
192 {
193 	struct parallel_io *par = bio->bi_private;
194 
195 	if (bio->bi_status) {
196 		struct nfs_pgio_header *header = par->data;
197 
198 		if (!header->pnfs_error)
199 			header->pnfs_error = -EIO;
200 		pnfs_set_lo_fail(header->lseg);
201 		bl_mark_devices_unavailable(header, false);
202 	}
203 
204 	bio_put(bio);
205 	put_parallel(par);
206 }
207 
208 static void bl_read_cleanup(struct work_struct *work)
209 {
210 	struct rpc_task *task;
211 	struct nfs_pgio_header *hdr;
212 	dprintk("%s enter\n", __func__);
213 	task = container_of(work, struct rpc_task, u.tk_work);
214 	hdr = container_of(task, struct nfs_pgio_header, task);
215 	pnfs_ld_read_done(hdr);
216 }
217 
218 static void
219 bl_end_par_io_read(void *data)
220 {
221 	struct nfs_pgio_header *hdr = data;
222 
223 	hdr->task.tk_status = hdr->pnfs_error;
224 	INIT_WORK(&hdr->task.u.tk_work, bl_read_cleanup);
225 	schedule_work(&hdr->task.u.tk_work);
226 }
227 
228 static enum pnfs_try_status
229 bl_read_pagelist(struct nfs_pgio_header *header)
230 {
231 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
232 	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
233 	struct bio *bio = NULL;
234 	struct pnfs_block_extent be;
235 	sector_t isect, extent_length = 0;
236 	struct parallel_io *par;
237 	loff_t f_offset = header->args.offset;
238 	size_t bytes_left = header->args.count;
239 	unsigned int pg_offset = header->args.pgbase, pg_len;
240 	struct page **pages = header->args.pages;
241 	int pg_index = header->args.pgbase >> PAGE_SHIFT;
242 	const bool is_dio = (header->dreq != NULL);
243 	struct blk_plug plug;
244 	int i;
245 
246 	dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
247 		header->page_array.npages, f_offset,
248 		(unsigned int)header->args.count);
249 
250 	par = alloc_parallel(header);
251 	if (!par)
252 		return PNFS_NOT_ATTEMPTED;
253 	par->pnfs_callback = bl_end_par_io_read;
254 
255 	blk_start_plug(&plug);
256 
257 	isect = (sector_t) (f_offset >> SECTOR_SHIFT);
258 	/* Code assumes extents are page-aligned */
259 	for (i = pg_index; i < header->page_array.npages; i++) {
260 		if (extent_length <= 0) {
261 			/* We've used up the previous extent */
262 			bio = bl_submit_bio(bio);
263 
264 			/* Get the next one */
265 			if (!ext_tree_lookup(bl, isect, &be, false)) {
266 				header->pnfs_error = -EIO;
267 				goto out;
268 			}
269 			extent_length = be.be_length - (isect - be.be_f_offset);
270 		}
271 
272 		if (is_dio) {
273 			if (pg_offset + bytes_left > PAGE_SIZE)
274 				pg_len = PAGE_SIZE - pg_offset;
275 			else
276 				pg_len = bytes_left;
277 		} else {
278 			BUG_ON(pg_offset != 0);
279 			pg_len = PAGE_SIZE;
280 		}
281 
282 		if (is_hole(&be)) {
283 			bio = bl_submit_bio(bio);
284 			/* Fill hole w/ zeroes w/o accessing device */
285 			dprintk("%s Zeroing page for hole\n", __func__);
286 			zero_user_segment(pages[i], pg_offset, pg_len);
287 
288 			/* invalidate map */
289 			map.start = NFS4_MAX_UINT64;
290 		} else {
291 			bio = do_add_page_to_bio(bio,
292 						 header->page_array.npages - i,
293 						 REQ_OP_READ,
294 						 isect, pages[i], &map, &be,
295 						 bl_end_io_read, par,
296 						 pg_offset, &pg_len);
297 			if (IS_ERR(bio)) {
298 				header->pnfs_error = PTR_ERR(bio);
299 				bio = NULL;
300 				goto out;
301 			}
302 		}
303 		isect += (pg_len >> SECTOR_SHIFT);
304 		extent_length -= (pg_len >> SECTOR_SHIFT);
305 		f_offset += pg_len;
306 		bytes_left -= pg_len;
307 		pg_offset = 0;
308 	}
309 	if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
310 		header->res.eof = 1;
311 		header->res.count = header->inode->i_size - header->args.offset;
312 	} else {
313 		header->res.count = (isect << SECTOR_SHIFT) - header->args.offset;
314 	}
315 out:
316 	bl_submit_bio(bio);
317 	blk_finish_plug(&plug);
318 	put_parallel(par);
319 	return PNFS_ATTEMPTED;
320 }
321 
322 static void bl_end_io_write(struct bio *bio)
323 {
324 	struct parallel_io *par = bio->bi_private;
325 	struct nfs_pgio_header *header = par->data;
326 
327 	if (bio->bi_status) {
328 		if (!header->pnfs_error)
329 			header->pnfs_error = -EIO;
330 		pnfs_set_lo_fail(header->lseg);
331 		bl_mark_devices_unavailable(header, true);
332 	}
333 	bio_put(bio);
334 	put_parallel(par);
335 }
336 
337 /* Function scheduled for call during bl_end_par_io_write,
338  * it marks sectors as written and extends the commitlist.
339  */
340 static void bl_write_cleanup(struct work_struct *work)
341 {
342 	struct rpc_task *task = container_of(work, struct rpc_task, u.tk_work);
343 	struct nfs_pgio_header *hdr =
344 			container_of(task, struct nfs_pgio_header, task);
345 
346 	dprintk("%s enter\n", __func__);
347 
348 	if (likely(!hdr->pnfs_error)) {
349 		struct pnfs_block_layout *bl = BLK_LSEG2EXT(hdr->lseg);
350 		u64 start = hdr->args.offset & (loff_t)PAGE_MASK;
351 		u64 end = (hdr->args.offset + hdr->args.count +
352 			PAGE_SIZE - 1) & (loff_t)PAGE_MASK;
353 		u64 lwb = hdr->args.offset + hdr->args.count;
354 
355 		ext_tree_mark_written(bl, start >> SECTOR_SHIFT,
356 					(end - start) >> SECTOR_SHIFT, lwb);
357 	}
358 
359 	pnfs_ld_write_done(hdr);
360 }
361 
362 /* Called when last of bios associated with a bl_write_pagelist call finishes */
363 static void bl_end_par_io_write(void *data)
364 {
365 	struct nfs_pgio_header *hdr = data;
366 
367 	hdr->task.tk_status = hdr->pnfs_error;
368 	hdr->verf.committed = NFS_FILE_SYNC;
369 	INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup);
370 	schedule_work(&hdr->task.u.tk_work);
371 }
372 
373 static enum pnfs_try_status
374 bl_write_pagelist(struct nfs_pgio_header *header, int sync)
375 {
376 	struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
377 	struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
378 	struct bio *bio = NULL;
379 	struct pnfs_block_extent be;
380 	sector_t isect, extent_length = 0;
381 	struct parallel_io *par = NULL;
382 	loff_t offset = header->args.offset;
383 	size_t count = header->args.count;
384 	struct page **pages = header->args.pages;
385 	int pg_index = header->args.pgbase >> PAGE_SHIFT;
386 	unsigned int pg_len;
387 	struct blk_plug plug;
388 	int i;
389 
390 	dprintk("%s enter, %zu@%lld\n", __func__, count, offset);
391 
392 	/* At this point, header->page_aray is a (sequential) list of nfs_pages.
393 	 * We want to write each, and if there is an error set pnfs_error
394 	 * to have it redone using nfs.
395 	 */
396 	par = alloc_parallel(header);
397 	if (!par)
398 		return PNFS_NOT_ATTEMPTED;
399 	par->pnfs_callback = bl_end_par_io_write;
400 
401 	blk_start_plug(&plug);
402 
403 	/* we always write out the whole page */
404 	offset = offset & (loff_t)PAGE_MASK;
405 	isect = offset >> SECTOR_SHIFT;
406 
407 	for (i = pg_index; i < header->page_array.npages; i++) {
408 		if (extent_length <= 0) {
409 			/* We've used up the previous extent */
410 			bio = bl_submit_bio(bio);
411 			/* Get the next one */
412 			if (!ext_tree_lookup(bl, isect, &be, true)) {
413 				header->pnfs_error = -EINVAL;
414 				goto out;
415 			}
416 
417 			extent_length = be.be_length - (isect - be.be_f_offset);
418 		}
419 
420 		pg_len = PAGE_SIZE;
421 		bio = do_add_page_to_bio(bio, header->page_array.npages - i,
422 					 REQ_OP_WRITE, isect, pages[i], &map,
423 					 &be, bl_end_io_write, par, 0, &pg_len);
424 		if (IS_ERR(bio)) {
425 			header->pnfs_error = PTR_ERR(bio);
426 			bio = NULL;
427 			goto out;
428 		}
429 
430 		offset += pg_len;
431 		count -= pg_len;
432 		isect += (pg_len >> SECTOR_SHIFT);
433 		extent_length -= (pg_len >> SECTOR_SHIFT);
434 	}
435 
436 	header->res.count = header->args.count;
437 out:
438 	bl_submit_bio(bio);
439 	blk_finish_plug(&plug);
440 	put_parallel(par);
441 	return PNFS_ATTEMPTED;
442 }
443 
444 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
445 {
446 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
447 	int err;
448 
449 	dprintk("%s enter\n", __func__);
450 
451 	err = ext_tree_remove(bl, true, 0, LLONG_MAX);
452 	WARN_ON(err);
453 
454 	kfree_rcu(bl, bl_layout.plh_rcu);
455 }
456 
457 static struct pnfs_layout_hdr *__bl_alloc_layout_hdr(struct inode *inode,
458 		gfp_t gfp_flags, bool is_scsi_layout)
459 {
460 	struct pnfs_block_layout *bl;
461 
462 	dprintk("%s enter\n", __func__);
463 	bl = kzalloc_obj(*bl, gfp_flags);
464 	if (!bl)
465 		return NULL;
466 
467 	bl->bl_ext_rw = RB_ROOT;
468 	bl->bl_ext_ro = RB_ROOT;
469 	spin_lock_init(&bl->bl_ext_lock);
470 
471 	bl->bl_scsi_layout = is_scsi_layout;
472 	return &bl->bl_layout;
473 }
474 
475 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
476 						   gfp_t gfp_flags)
477 {
478 	return __bl_alloc_layout_hdr(inode, gfp_flags, false);
479 }
480 
481 static struct pnfs_layout_hdr *sl_alloc_layout_hdr(struct inode *inode,
482 						   gfp_t gfp_flags)
483 {
484 	return __bl_alloc_layout_hdr(inode, gfp_flags, true);
485 }
486 
487 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
488 {
489 	dprintk("%s enter\n", __func__);
490 	kfree(lseg);
491 }
492 
493 /* Tracks info needed to ensure extents in layout obey constraints of spec */
494 struct layout_verification {
495 	u32 mode;	/* R or RW */
496 	u64 start;	/* Expected start of next non-COW extent */
497 	u64 inval;	/* Start of INVAL coverage */
498 	u64 cowread;	/* End of COW read coverage */
499 };
500 
501 /* Verify the extent meets the layout requirements of the pnfs-block draft,
502  * section 2.3.1.
503  */
504 static int verify_extent(struct pnfs_block_extent *be,
505 			 struct layout_verification *lv)
506 {
507 	if (lv->mode == IOMODE_READ) {
508 		if (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
509 		    be->be_state == PNFS_BLOCK_INVALID_DATA)
510 			return -EIO;
511 		if (be->be_f_offset != lv->start)
512 			return -EIO;
513 		lv->start += be->be_length;
514 		return 0;
515 	}
516 	/* lv->mode == IOMODE_RW */
517 	if (be->be_state == PNFS_BLOCK_READWRITE_DATA) {
518 		if (be->be_f_offset != lv->start)
519 			return -EIO;
520 		if (lv->cowread > lv->start)
521 			return -EIO;
522 		lv->start += be->be_length;
523 		lv->inval = lv->start;
524 		return 0;
525 	} else if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
526 		if (be->be_f_offset != lv->start)
527 			return -EIO;
528 		lv->start += be->be_length;
529 		return 0;
530 	} else if (be->be_state == PNFS_BLOCK_READ_DATA) {
531 		if (be->be_f_offset > lv->start)
532 			return -EIO;
533 		if (be->be_f_offset < lv->inval)
534 			return -EIO;
535 		if (be->be_f_offset < lv->cowread)
536 			return -EIO;
537 		/* It looks like you might want to min this with lv->start,
538 		 * but you really don't.
539 		 */
540 		lv->inval = lv->inval + be->be_length;
541 		lv->cowread = be->be_f_offset + be->be_length;
542 		return 0;
543 	} else
544 		return -EIO;
545 }
546 
547 static int decode_sector_number(__be32 **rp, sector_t *sp)
548 {
549 	uint64_t s;
550 
551 	*rp = xdr_decode_hyper(*rp, &s);
552 	if (s & 0x1ff) {
553 		printk(KERN_WARNING "NFS: %s: sector not aligned\n", __func__);
554 		return -1;
555 	}
556 	*sp = s >> SECTOR_SHIFT;
557 	return 0;
558 }
559 
560 static struct nfs4_deviceid_node *
561 bl_find_get_deviceid(struct nfs_server *server,
562 		const struct nfs4_deviceid *id, const struct cred *cred,
563 		gfp_t gfp_mask)
564 {
565 	struct nfs4_deviceid_node *node;
566 	int err = -ENODEV;
567 
568 retry:
569 	node = nfs4_find_get_deviceid(server, id, cred, gfp_mask);
570 	if (!node)
571 		return ERR_PTR(-ENODEV);
572 
573 	/*
574 	 * Devices that are marked unavailable are left in the cache with a
575 	 * timeout to avoid sending GETDEVINFO after every LAYOUTGET, or
576 	 * constantly attempting to register the device.  Once marked as
577 	 * unavailable they must be deleted and never reused.
578 	 */
579 	if (test_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags)) {
580 		unsigned long end = jiffies;
581 		unsigned long start = end - PNFS_DEVICE_RETRY_TIMEOUT;
582 
583 		if (!time_in_range(node->timestamp_unavailable, start, end)) {
584 			/* Uncork subsequent GETDEVINFO operations for this device */
585 			nfs4_delete_deviceid(node->ld, node->nfs_client, id);
586 			goto retry;
587 		}
588 		goto out_put;
589 	}
590 
591 	if (!bl_register_dev(container_of(node, struct pnfs_block_dev, node))) {
592 		/*
593 		 * If we cannot register, treat this device as transient:
594 		 * Make a negative cache entry for the device
595 		 */
596 		nfs4_mark_deviceid_unavailable(node);
597 		goto out_put;
598 	}
599 
600 	return node;
601 
602 out_put:
603 	nfs4_put_deviceid_node(node);
604 	return ERR_PTR(err);
605 }
606 
607 static int
608 bl_alloc_extent(struct xdr_stream *xdr, struct pnfs_layout_hdr *lo,
609 		struct layout_verification *lv, struct list_head *extents,
610 		gfp_t gfp_mask)
611 {
612 	struct pnfs_block_extent *be;
613 	struct nfs4_deviceid id;
614 	int error;
615 	__be32 *p;
616 
617 	p = xdr_inline_decode(xdr, 28 + NFS4_DEVICEID4_SIZE);
618 	if (!p)
619 		return -EIO;
620 
621 	be = kzalloc_obj(*be, GFP_NOFS);
622 	if (!be)
623 		return -ENOMEM;
624 
625 	memcpy(&id, p, NFS4_DEVICEID4_SIZE);
626 	p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
627 
628 	be->be_device = bl_find_get_deviceid(NFS_SERVER(lo->plh_inode), &id,
629 						lo->plh_lc_cred, gfp_mask);
630 	if (IS_ERR(be->be_device)) {
631 		error = PTR_ERR(be->be_device);
632 		goto out_free_be;
633 	}
634 
635 	/*
636 	 * The next three values are read in as bytes, but stored in the
637 	 * extent structure in 512-byte granularity.
638 	 */
639 	error = -EIO;
640 	if (decode_sector_number(&p, &be->be_f_offset) < 0)
641 		goto out_put_deviceid;
642 	if (decode_sector_number(&p, &be->be_length) < 0)
643 		goto out_put_deviceid;
644 	if (decode_sector_number(&p, &be->be_v_offset) < 0)
645 		goto out_put_deviceid;
646 	be->be_state = be32_to_cpup(p++);
647 
648 	error = verify_extent(be, lv);
649 	if (error) {
650 		dprintk("%s: extent verification failed\n", __func__);
651 		goto out_put_deviceid;
652 	}
653 
654 	list_add_tail(&be->be_list, extents);
655 	return 0;
656 
657 out_put_deviceid:
658 	nfs4_put_deviceid_node(be->be_device);
659 out_free_be:
660 	kfree(be);
661 	return error;
662 }
663 
664 static struct pnfs_layout_segment *
665 bl_alloc_lseg(struct pnfs_layout_hdr *lo, struct nfs4_layoutget_res *lgr,
666 		gfp_t gfp_mask)
667 {
668 	struct layout_verification lv = {
669 		.mode = lgr->range.iomode,
670 		.start = lgr->range.offset >> SECTOR_SHIFT,
671 		.inval = lgr->range.offset >> SECTOR_SHIFT,
672 		.cowread = lgr->range.offset >> SECTOR_SHIFT,
673 	};
674 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
675 	struct pnfs_layout_segment *lseg;
676 	struct xdr_buf buf;
677 	struct xdr_stream xdr;
678 	struct folio *scratch;
679 	int status, i;
680 	uint32_t count;
681 	__be32 *p;
682 	LIST_HEAD(extents);
683 
684 	dprintk("---> %s\n", __func__);
685 
686 	lseg = kzalloc_obj(*lseg, gfp_mask);
687 	if (!lseg)
688 		return ERR_PTR(-ENOMEM);
689 
690 	status = -ENOMEM;
691 	scratch = folio_alloc(gfp_mask, 0);
692 	if (!scratch)
693 		goto out;
694 
695 	xdr_init_decode_pages(&xdr, &buf,
696 			lgr->layoutp->pages, lgr->layoutp->len);
697 	xdr_set_scratch_folio(&xdr, scratch);
698 
699 	status = -EIO;
700 	p = xdr_inline_decode(&xdr, 4);
701 	if (unlikely(!p))
702 		goto out_free_scratch;
703 
704 	count = be32_to_cpup(p++);
705 	dprintk("%s: number of extents %d\n", __func__, count);
706 
707 	/*
708 	 * Decode individual extents, putting them in temporary staging area
709 	 * until whole layout is decoded to make error recovery easier.
710 	 */
711 	for (i = 0; i < count; i++) {
712 		status = bl_alloc_extent(&xdr, lo, &lv, &extents, gfp_mask);
713 		if (status)
714 			goto process_extents;
715 	}
716 
717 	if (lgr->range.offset + lgr->range.length !=
718 			lv.start << SECTOR_SHIFT) {
719 		dprintk("%s Final length mismatch\n", __func__);
720 		status = -EIO;
721 		goto process_extents;
722 	}
723 
724 	if (lv.start < lv.cowread) {
725 		dprintk("%s Final uncovered COW extent\n", __func__);
726 		status = -EIO;
727 	}
728 
729 process_extents:
730 	while (!list_empty(&extents)) {
731 		struct pnfs_block_extent *be =
732 			list_first_entry(&extents, struct pnfs_block_extent,
733 					 be_list);
734 		list_del(&be->be_list);
735 
736 		if (!status)
737 			status = ext_tree_insert(bl, be);
738 
739 		if (status) {
740 			nfs4_put_deviceid_node(be->be_device);
741 			kfree(be);
742 		}
743 	}
744 
745 out_free_scratch:
746 	folio_put(scratch);
747 out:
748 	dprintk("%s returns %d\n", __func__, status);
749 	switch (status) {
750 	case -ENODEV:
751 		/* Our extent block devices are unavailable */
752 		set_bit(NFS_LSEG_UNAVAILABLE, &lseg->pls_flags);
753 		fallthrough;
754 	case 0:
755 		return lseg;
756 	default:
757 		kfree(lseg);
758 		return ERR_PTR(status);
759 	}
760 }
761 
762 static void
763 bl_return_range(struct pnfs_layout_hdr *lo,
764 		struct pnfs_layout_range *range)
765 {
766 	struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
767 	sector_t offset = range->offset >> SECTOR_SHIFT, end;
768 
769 	if (range->offset % 8) {
770 		dprintk("%s: offset %lld not block size aligned\n",
771 			__func__, range->offset);
772 		return;
773 	}
774 
775 	if (range->length != NFS4_MAX_UINT64) {
776 		if (range->length % 8) {
777 			dprintk("%s: length %lld not block size aligned\n",
778 				__func__, range->length);
779 			return;
780 		}
781 
782 		end = offset + (range->length >> SECTOR_SHIFT);
783 	} else {
784 		end = round_down(NFS4_MAX_UINT64, PAGE_SIZE);
785 	}
786 
787 	ext_tree_remove(bl, range->iomode & IOMODE_RW, offset, end);
788 }
789 
790 static int
791 bl_prepare_layoutcommit(struct nfs4_layoutcommit_args *arg)
792 {
793 	return ext_tree_prepare_commit(arg);
794 }
795 
796 static void
797 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
798 {
799 	ext_tree_mark_committed(&lcdata->args, lcdata->res.status);
800 }
801 
802 static int
803 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
804 {
805 	dprintk("%s enter\n", __func__);
806 
807 	if (server->pnfs_blksize == 0) {
808 		dprintk("%s Server did not return blksize\n", __func__);
809 		return -EINVAL;
810 	}
811 	if (server->pnfs_blksize > PAGE_SIZE) {
812 		printk(KERN_ERR "%s: pNFS blksize %d not supported.\n",
813 			__func__, server->pnfs_blksize);
814 		return -EINVAL;
815 	}
816 
817 	return 0;
818 }
819 
820 static bool
821 is_aligned_req(struct nfs_pageio_descriptor *pgio,
822 		struct nfs_page *req, unsigned int alignment, bool is_write)
823 {
824 	/*
825 	 * Always accept buffered writes, higher layers take care of the
826 	 * right alignment.
827 	 */
828 	if (pgio->pg_dreq == NULL)
829 		return true;
830 
831 	if (!IS_ALIGNED(req->wb_offset, alignment))
832 		return false;
833 
834 	if (IS_ALIGNED(req->wb_bytes, alignment))
835 		return true;
836 
837 	if (is_write &&
838 	    (req_offset(req) + req->wb_bytes == i_size_read(pgio->pg_inode))) {
839 		/*
840 		 * If the write goes up to the inode size, just write
841 		 * the full page.  Data past the inode size is
842 		 * guaranteed to be zeroed by the higher level client
843 		 * code, and this behaviour is mandated by RFC 5663
844 		 * section 2.3.2.
845 		 */
846 		return true;
847 	}
848 
849 	return false;
850 }
851 
852 static void
853 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
854 {
855 	if (!is_aligned_req(pgio, req, SECTOR_SIZE, false)) {
856 		nfs_pageio_reset_read_mds(pgio);
857 		return;
858 	}
859 
860 	pnfs_generic_pg_init_read(pgio, req);
861 
862 	if (pgio->pg_lseg &&
863 		test_bit(NFS_LSEG_UNAVAILABLE, &pgio->pg_lseg->pls_flags)) {
864 		pnfs_error_mark_layout_for_return(pgio->pg_inode, pgio->pg_lseg);
865 		pnfs_set_lo_fail(pgio->pg_lseg);
866 		nfs_pageio_reset_read_mds(pgio);
867 	}
868 }
869 
870 /*
871  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
872  * of bytes (maximum @req->wb_bytes) that can be coalesced.
873  */
874 static size_t
875 bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
876 		struct nfs_page *req)
877 {
878 	if (!is_aligned_req(pgio, req, SECTOR_SIZE, false))
879 		return 0;
880 	return pnfs_generic_pg_test(pgio, prev, req);
881 }
882 
883 /*
884  * Return the number of contiguous bytes for a given inode
885  * starting at page frame idx.
886  */
887 static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx)
888 {
889 	struct address_space *mapping = inode->i_mapping;
890 	pgoff_t end;
891 
892 	/* Optimize common case that writes from 0 to end of file */
893 	end = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
894 	if (end != inode->i_mapping->nrpages) {
895 		rcu_read_lock();
896 		end = page_cache_next_miss(mapping, idx + 1, ULONG_MAX);
897 		rcu_read_unlock();
898 	}
899 
900 	if (!end)
901 		return i_size_read(inode) - (idx << PAGE_SHIFT);
902 	else
903 		return (end - idx) << PAGE_SHIFT;
904 }
905 
906 static void
907 bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
908 {
909 	u64 wb_size;
910 
911 	if (!is_aligned_req(pgio, req, PAGE_SIZE, true)) {
912 		nfs_pageio_reset_write_mds(pgio);
913 		return;
914 	}
915 
916 	if (pgio->pg_dreq == NULL)
917 		wb_size = pnfs_num_cont_bytes(pgio->pg_inode, req->wb_index);
918 	else
919 		wb_size = nfs_dreq_bytes_left(pgio->pg_dreq, req_offset(req));
920 
921 	pnfs_generic_pg_init_write(pgio, req, wb_size);
922 
923 	if (pgio->pg_lseg &&
924 		test_bit(NFS_LSEG_UNAVAILABLE, &pgio->pg_lseg->pls_flags)) {
925 
926 		pnfs_error_mark_layout_for_return(pgio->pg_inode, pgio->pg_lseg);
927 		pnfs_set_lo_fail(pgio->pg_lseg);
928 		nfs_pageio_reset_write_mds(pgio);
929 	}
930 }
931 
932 /*
933  * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
934  * of bytes (maximum @req->wb_bytes) that can be coalesced.
935  */
936 static size_t
937 bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
938 		 struct nfs_page *req)
939 {
940 	if (!is_aligned_req(pgio, req, PAGE_SIZE, true))
941 		return 0;
942 	return pnfs_generic_pg_test(pgio, prev, req);
943 }
944 
945 static const struct nfs_pageio_ops bl_pg_read_ops = {
946 	.pg_init = bl_pg_init_read,
947 	.pg_test = bl_pg_test_read,
948 	.pg_doio = pnfs_generic_pg_readpages,
949 	.pg_cleanup = pnfs_generic_pg_cleanup,
950 };
951 
952 static const struct nfs_pageio_ops bl_pg_write_ops = {
953 	.pg_init = bl_pg_init_write,
954 	.pg_test = bl_pg_test_write,
955 	.pg_doio = pnfs_generic_pg_writepages,
956 	.pg_cleanup = pnfs_generic_pg_cleanup,
957 };
958 
959 static struct pnfs_layoutdriver_type blocklayout_type = {
960 	.id				= LAYOUT_BLOCK_VOLUME,
961 	.name				= "LAYOUT_BLOCK_VOLUME",
962 	.owner				= THIS_MODULE,
963 	.flags				= PNFS_LAYOUTRET_ON_SETATTR |
964 					  PNFS_LAYOUTRET_ON_ERROR |
965 					  PNFS_READ_WHOLE_PAGE,
966 	.read_pagelist			= bl_read_pagelist,
967 	.write_pagelist			= bl_write_pagelist,
968 	.alloc_layout_hdr		= bl_alloc_layout_hdr,
969 	.free_layout_hdr		= bl_free_layout_hdr,
970 	.alloc_lseg			= bl_alloc_lseg,
971 	.free_lseg			= bl_free_lseg,
972 	.return_range			= bl_return_range,
973 	.prepare_layoutcommit		= bl_prepare_layoutcommit,
974 	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
975 	.set_layoutdriver		= bl_set_layoutdriver,
976 	.alloc_deviceid_node		= bl_alloc_deviceid_node,
977 	.free_deviceid_node		= bl_free_deviceid_node,
978 	.pg_read_ops			= &bl_pg_read_ops,
979 	.pg_write_ops			= &bl_pg_write_ops,
980 	.sync				= pnfs_generic_sync,
981 };
982 
983 static struct pnfs_layoutdriver_type scsilayout_type = {
984 	.id				= LAYOUT_SCSI,
985 	.name				= "LAYOUT_SCSI",
986 	.owner				= THIS_MODULE,
987 	.flags				= PNFS_LAYOUTRET_ON_SETATTR |
988 					  PNFS_LAYOUTRET_ON_ERROR |
989 					  PNFS_READ_WHOLE_PAGE,
990 	.read_pagelist			= bl_read_pagelist,
991 	.write_pagelist			= bl_write_pagelist,
992 	.alloc_layout_hdr		= sl_alloc_layout_hdr,
993 	.free_layout_hdr		= bl_free_layout_hdr,
994 	.alloc_lseg			= bl_alloc_lseg,
995 	.free_lseg			= bl_free_lseg,
996 	.return_range			= bl_return_range,
997 	.prepare_layoutcommit		= bl_prepare_layoutcommit,
998 	.cleanup_layoutcommit		= bl_cleanup_layoutcommit,
999 	.set_layoutdriver		= bl_set_layoutdriver,
1000 	.alloc_deviceid_node		= bl_alloc_deviceid_node,
1001 	.free_deviceid_node		= bl_free_deviceid_node,
1002 	.pg_read_ops			= &bl_pg_read_ops,
1003 	.pg_write_ops			= &bl_pg_write_ops,
1004 	.sync				= pnfs_generic_sync,
1005 };
1006 
1007 
1008 static int __init nfs4blocklayout_init(void)
1009 {
1010 	int ret;
1011 
1012 	dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
1013 
1014 	ret = bl_init_pipefs();
1015 	if (ret)
1016 		goto out;
1017 
1018 	ret = pnfs_register_layoutdriver(&blocklayout_type);
1019 	if (ret)
1020 		goto out_cleanup_pipe;
1021 
1022 	ret = pnfs_register_layoutdriver(&scsilayout_type);
1023 	if (ret)
1024 		goto out_unregister_block;
1025 	return 0;
1026 
1027 out_unregister_block:
1028 	pnfs_unregister_layoutdriver(&blocklayout_type);
1029 out_cleanup_pipe:
1030 	bl_cleanup_pipefs();
1031 out:
1032 	return ret;
1033 }
1034 
1035 static void __exit nfs4blocklayout_exit(void)
1036 {
1037 	dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
1038 	       __func__);
1039 
1040 	pnfs_unregister_layoutdriver(&scsilayout_type);
1041 	pnfs_unregister_layoutdriver(&blocklayout_type);
1042 	bl_cleanup_pipefs();
1043 }
1044 
1045 MODULE_ALIAS("nfs-layouttype4-3");
1046 MODULE_ALIAS("nfs-layouttype4-5");
1047 
1048 module_init(nfs4blocklayout_init);
1049 module_exit(nfs4blocklayout_exit);
1050